With excellent compression rate and fine scalability, an inter-frame wavelet coding method is generally appropriate for diverse application apparatuses that require a function of scalability. The video scalability is a core function that secures quality of service (QoS) in a heterogeneous network environment such as a network integrating broadcasting and communication.
Since a conventional video coding method based on DCT (DCT) is optimized for supporting a single scalability function, it is not suitable to a service requiring a multiple scalability function such as a heterogeneous network environment.
Inter-frame wavelet coding is also called three-dimensional (3D) sub-band coding.
This is because two-dimensional video wavelet transform is performed on video which is obtained by performing sub-band filtering with respect to the temporal axis.
When filtering is performed with respect to the temporal axis, the filtering is not a simple filtering but it is performed in consideration of the motion of video. This filtering is referred to as Motion Compensated Temporal filtering (MCTF).
The inter-frame wavelet coding begins with performing MCTF on video inputted on a basis of group of picture (GOP).
MCTF includes two methods: One is a method of filtering a plurality of frames with respect to the temporal axis, and the other is a method of filtering two frames with respect to the temporal axis. According to the method of filtering a plurality of frames with respect to the temporal axis, original frames are set as reference frames and highpass-filtered frames are generated based on the reference frames and their neighboring frames.
According to the method of filtering two frames with respect to the temporal axis, lowpass-filtered and highpass-filtered frames are generated with reference to the two neighboring frames.
Here, a motion vector for each block is calculated with respect to input video prior to being filtered and the calculated motion vector is applied to filtering.
The methods for applying the motion vector to MCTF include an Ohm method and a Woods method. Generally, the Woods method is known and used for higher efficiency in calculating a motion vector.
Meanwhile, a Haar filter is widely used. The Haar filter performs lowpass-filtering and highpass-filtering by regarding two neighboring frames as one unit.
FIG. 1 is a diagram illustrating a motion compensated temporal filtering (MCTF) process using the Haar filter. For convenience in description, it is assumed that the GOP is 8 in the MCTF process of FIG. 1.
As shown, when MCTF is performed on the two neighboring frames, four t-L frames which are lowpass-filtered on the temporal axis and four t-H frames 111, 112, 113 and 114 which are highpass-filtered on the temporal axis are generated. The eight frames are called a first temporal level 110.
When the MCTF is performed again on the four t-L lowpass-filtered frames, two t-LL lowpass-filtered frames and two t-LH highpass-filtered frames 121 and 122 are generated. The four frames are called a second temporal level 120.
If the MCTF is performed again on the two t-LL lowpass-filtered frames, a t-LLL frame 131 and a t-LLH frame 132 are generated. The two frames 131 and 132 are called a third temporal level 130.
The eight frames 111, 112, 113, 114, 121, 122, 131 and 132 on which the MCTF is performed go through spatial wavelet transform, quantization and entropy coding and eventually, bit stream is generated and transmitted to a receiving end.
Conventionally, a 9/7 filter or a 5/3 filter is used uniformly in a step of spatial wavelet transform, regardless of characteristics of a frame.
If the wavelet transform is carried out using all the same filters regardless of characteristics of a frame, there is a problem that the amount of calculation is increased unnecessarily during inverse wavelet transform of decoding. The amount of calculation required for the inverse wavelet transform is increased as the number of decomposition levels of wavelet transform is larger and the length of a filter is longer.
If the calculation amount of the inverse wavelet transform is increased the inter-frame wavelet coding method cannot be used in a terminal having low computing capability such as a Personal Digital Assistant (PDA).